Parth Asawa
Bio
Parth Asawa is a PhD student at UC Berkeley advised by Professor Matei Zaharia and Professor Joey Gonzalez. Parth's research is on continual learning, studying how to enable models to stably learn from streams of experiences over time. His work focuses on sample-efficient learning and spans the stack of data, learning algorithms, and evaluation.
Session (1)
Parth Asawa is a Berkeley CS PhD student working at the frontier of continual learning — building benchmarks, sample-efficient learning algorithms, and advisor-model techniques that let AI systems genuinely improve from experience rather than just at training time. Attending his session offers a grounded, systems-level perspective on what it actually takes to make LLM-based agents learn over time, backed by recent papers at VLDB, ICML, and NeurIPS.
GitHub
@pgasawa- pgasawa/continual-learning-bench★ 162Benchmark for measuring how well AI agents learn from past environment interactions across six real-world domains.
- pgasawa/BARE★ 41Leverages base language models for few-shot synthetic data generation, achieving up to 18.4% improvement over SOTA on RAFT tasks.
- pgasawa/SIEVE★ 11Sample-efficient parametric learning from natural language, distilling context into model weights using as few as 3 query examples.
Recent writing (5)
- Unsafe AI or Consolidated Power: AI's False Dichotomy · blog · Jun 2026
- Continual Learning Bench: Evaluating Frontier AI Systems in Real-World Stateful Environments · paper · Jun 2026
- SIEVE: Sample-Efficient Parametric Learning from Natural Language · paper · Feb 2026
- How to Train Your Advisor: Steering Black-Box LLMs with Advisor Models · paper · Oct 2025
- Semantic Operators and Their Optimization: Enabling LLM-Based Data Processing with Accuracy Guarantees in LOTUS · paper · Sep 2025
Public activity researched automatically · as of Jun 2026